A Natural Experiment in Fiscal Expectations
Government debt and long-term Treasury rates meet most clearly when markets must reprice a government’s future fiscal path. This chapter focuses on the 2021 Georgia Senate runoff, the episode a Federal Reserve study uses to isolate that repricing from the usual background movement in interest rates. Following the episode from political uncertainty to expected debt, market prices and statistical checks shows what a causal estimate can reveal—and why it should not be treated as a universal debt-to-rate rule.
The political branch markets had not resolved
By the time Georgia voters went to the polls in January 2021, much of the political outlook had already been incorporated into financial prices. The November 2020 election had settled the presidency and the House majority. Two Senate seats in Georgia remained undecided, however, and together they determined whether the incoming administration would control the Senate.
That distinction mattered because the Senate majority would not have given the administration unlimited legislative power. Ordinary legislation could still be blocked by the filibuster. Fiscal policy passed through budget reconciliation was different: it required only a simple majority. A Democratic victory in both Georgia runoffs therefore created a specific policy branch that a divided government did not. The likely near-term difference was an additional, debt-financed fiscal expansion rather than a wholly different set of economic policies.
In December 2020, Congress had already approved roughly $900 billion in additional pandemic stimulus. The political debate then focused on whether that support would be increased, including through larger direct payments. Investment banks following the election reportedly expected about $900 billion in additional stimulus if Democrats gained control of the Senate and no comparable package under a divided government. They expected most of the money to take the form of transfers and almost all of it to be financed through new debt rather than taxes.
The relevant information for an asset price was not the final amount of legislation. It was the difference between the two policy branches, multiplied by the probability that each branch would occur. On the eve of the runoff, prediction markets put the chance of a Democratic sweep near 50 percent. A simplified representation of the expected change is therefore:
change in expected deficit ≈ change in probability × conditional fiscal package
≈ 50 percentage points × $900 billion = about $450 billion.
The calculation is an expectation, not a historical claim that the government immediately borrowed exactly $450 billion. Before the result, markets were pricing a chance of the larger package; after the result, that chance moved close to certainty. The repricing of the probability was the shock. This is why the episode is useful for studying long-term interest rates: investors could revise their view of future Treasury issuance quickly, before the government had to issue the securities or collect the transfers.
The result arrived in stages. News confirmed Raphael Warnock’s victory early on January 6, while the other race was called later. Betting markets nevertheless moved toward an almost certain Democratic victory during the early morning. Within one day, the 10-year Treasury yield rose by roughly 10 basis points. In hindsight, it is tempting to read that movement as a clean fiscal experiment. At the time, though, several things were happening around the same date. The study must therefore establish why the runoff result, rather than some other event, is the relevant source of the rate movement.
The case is not a story about a single political decision-maker changing rates by declaration. It is a story about a conditional legislative possibility becoming more likely, and about investors repricing the government debt that could follow. That difference keeps the economic mechanism in view and avoids attributing more information to the event than markets actually had.
How the study turns the episode into evidence
The exclusion restriction
The central identifying assumption is an exclusion restriction: the runoff outcome affected longer-term interest rates primarily through its effect on expected federal debt, rather than through an unrelated change in monetary-policy expectations, productivity, population growth or the near-term macroeconomic outlook.
The authors give several reasons for treating that assumption as plausible in this setting. The presidency and House majority had already been resolved in November and were largely priced soon afterward. The runoff concerned two Senate seats, not a change in control of the executive branch. And because the filibuster limited the scope of a narrow Senate majority for non-budgetary legislation, the most immediate policy implication of unified control was the possibility of fiscal action through reconciliation. Contemporary market commentary and investment-bank analysis also focused mainly on the fiscal consequences.
The timing helps as well. Slow-moving forces such as productivity and demographic trends are unlikely to have changed sharply between one trading session and the next. An overnight repricing of a long-term Treasury yield is therefore less naturally explained by those forces than by a discrete change in policy expectations. That is supportive evidence, not a logical guarantee. An election can carry information about more than one policy channel, and a single event cannot reveal every unobserved influence with certainty.
The exclusion restriction is best understood as a comparison between plausible explanations, not as a claim that the event occurred in a vacuum. The study asks whether the fiscal channel is sufficiently dominant and sufficiently well timed to identify the effect of expected debt. If the runoff had simultaneously produced a major revision to expected monetary policy or long-run growth, the rate movement could not be assigned to debt supply alone. The usefulness of the case comes from narrowing those alternatives, not from making them disappear.
From the yield reaction to an elasticity
A one-day change in the 10-year yield is not yet the final result. The researchers first measure the movements in nominal Treasury yields and real TIPS yields, then use a term-structure model to separate real yield changes into expected longer-run real rates and real term premiums. They scale those movements by the estimated size of the expected deficit shock, approximately 2 percent of GDP, so the reported coefficient answers a standardized question: how many basis points would the relevant rate change for a one-percentage-point increase in the expected debt-to-GDP ratio?
This unit matters. The study is not saying that a 2 percent-of-GDP fiscal shock produces a 2 percent increase in yields. Nor is it saying that a government with a debt ratio 1 percentage point higher automatically pays a fixed number of basis points on every bond. The reported number is an elasticity derived from an event whose expected fiscal effect was estimated relative to GDP.
The market reaction is economically visible but modest compared with the size of the headline yield move. The 10-year Treasury yield rose about 10 basis points around the resolution of the races, while the study’s scaled estimate for a one-percentage-point debt-to-GDP shock is a roughly 3–4 basis-point increase in the five-year-forward 10-year real yield. The difference reflects the scale of the episode: the expected debt change was about two percentage points of GDP, and the study focuses on the part of the yield movement attributable to that shock and its components.
The decomposition is central to the interpretation. For the one-day event window, the paper estimates a 3.96-basis-point increase in the five-year-forward 10-year TIPS yield per one-percentage-point debt-to-GDP shock. About 1.64 basis points are attributed to expected r*, the longer-run neutral real rate, and about 2.61 basis points to the real term premium. A narrower window ending at 12:50 p.m. on January 6 produces similar figures: 3.80 basis points for the real yield, 1.28 for expected r* and 2.52 for the real term premium.
The headline lesson is not that the neutral rate is irrelevant. It is that the larger estimated response in this event came from the compensation investors required to hold longer-term government debt. That result will matter later when we ask how additional Treasury duration is absorbed. For now, it establishes the object to be explained: the debt shock moved long rates, and the term premium appears to have carried most of the movement.
What robustness checks change
A concern about the full daily window is that yields could also have responded to developments surrounding the January 6 Capitol riots. The study addresses this by examining higher-frequency, five-minute yield changes before 12:50 p.m., before news of the riots had been reported. The estimated elasticities and the decomposition remain nearly identical. That makes it less likely that the main result is an artifact of later political violence, although it does not prove that every minute in the window contained only fiscal information.
The authors also use the probability of Democratic Senate control as a continuous measure during the weeks leading up to the runoff. Rather than relying only on the final outcome, they relate daily changes in Treasury yields and their model-based components to changes in the end-of-day probability. After rescaling the probability movement to represent a one-percentage-point debt-to-GDP shock, the estimates again put the expected r* response around 1–2 basis points and the real term-premium response around 2–3 basis points. A similar result from gradual probability changes supports the interpretation that the market was responding to fiscal expectations rather than only to a single news headline.
Placebo comparisons provide another check. The study constructs a distribution of yield responses from earlier dates and asks how often a shock of the relevant size would produce a movement as large as the January 6 estimate by chance. The reported daily components are statistically significant against that placebo distribution. This supports the claim that the event was unusual relative to ordinary daily movements, but statistical significance does not solve the external-validity problem. An unusual US event can identify a local effect without telling us exactly how a different government, market or monetary regime would respond.
That limit is not a technical footnote. The expected stimulus was primarily a one-time transfer package in the United States during a particular period. Investors’ response could differ when debt is issued at other maturities, when inflation credibility is weaker, when foreign investors dominate demand or when financial intermediaries are already under stress. The same coefficient also need not describe a change in actual debt after legislation is enacted, because the study identifies the effect of a revision in expectations.
The bounded lesson of the case
The Georgia runoff provides a comparatively clean answer to a difficult question: when investors revise the expected supply of US government debt, long-term real yields rise, and the term premium accounts for most of the measured response in this episode. The political design, rapid market repricing, narrow intraday check and probability-based regression make that conclusion more credible than a simple historical correlation.
But the episode does not show that debt always raises rates by a fixed amount, that every yield increase is fiscal in origin or that a higher term premium is the same thing as default risk. It also does not settle the full mechanism. The event study identifies the response; its model-based decomposition indicates where that response appeared. To understand why compensation for holding duration moved more than the long-run neutral rate—and why the same response might be larger in a fragile market—we need to examine how investors absorb government bonds and how debt supply interacts with their balance sheets.
That is the useful handoff from this case. A well-identified event gives us a measured yield response, but the number becomes economically meaningful only when we can explain the term premium that sits inside it.